Preliminary Results Assessing Observation Impact Within RRFS Using an Observation-Space Metric

In currently operational and developing regional convection-allowing weather forecast systems, millions of observations are assimilated on a sub-daily timescale. At NOAA, these systems include the High-Resolution Rapid Refresh (HRRR) v4 and the Rapid Fresh Forecast System (RRFS). However, up to date, there is a lack of a flexible and efficient tool to estimate the observation impact on analyses and short-range forecasts. Typically, to understand the impact of specific data type(s), one often performs the so-called data-denial or data add-on experiments, where a full retrospective experiment with the assimilation of all available observations and then another identical experiment except adding or denying those specified datasets. These experiments are not only computationally expensive but their results are constrained by the relative impact of the added or denied observations.

To address this, we employ a simple observation space-based metric to quantify the data impact for RRFS. This approach utilizes the background and analysis residuals usually readily available from data assimilation systems, and it relies on the forecast error reduction verified at the analysis time. Preliminary results obtained from recent RRFS real-time runs show that the total impact from satellite data and that from conventional data are comparable, although the observation counts of assimilated satellite data are relatively larger than the observation counts of assimilated conventional data. Among various satellite sensors, AMSU-A shows a relatively large impact per observation, while the infrared data on the Cross-track Infrared Sounder (CrIS) and the Infrared Atmospheric Sound Interferometer (IASI) show a strong total impact, due to the large size of assimilated infrared data.